| name | anypoc |
| description | AnyPoC skills — setup projects, find bugs, generate PoCs, retry a PoC run with new guidance, hunt (scan + PoC in one pass), launch the dashboard, check status of active runs, or list all subcommands via help. Subcommands include setup-project, find-bugs, run-poc, retry-poc, hunt, dashboard, status, help. |
| disable-model-invocation | true |
| argument-hint | hunt PROJECT for what kinds of bugs |
You are running an AnyPoC skill. The subcommand is "$0" with arguments: $1
Provider note (applies to hunt, find-bugs / scan, run-poc, retry-poc): these commands spawn a coding agent inside each container via caw. caw picks the agent from the CAW_PROVIDER env var (defaults to claude). If you yourself are running as Codex (not Claude Code), prepend CAW_PROVIDER=codex to every anypoc command you build for the user — both the raw command and the tmux-wrapped variant — so the in-container agent matches the host agent the user is already using and authenticated with. Example: CAW_PROVIDER=codex anypoc hunt run history -p wasmtime time_range="last 6 months". Skip this for dashboard, status, setup-project, and help — they don't spawn agents.
Model note: CAW_MODEL selects which model the in-container agent runs with. Leave it unset to use caw's default for the chosen provider — that is the right choice for almost all runs. Only set it (e.g. CAW_MODEL=claude-opus-4-7 anypoc hunt run ...) if the user explicitly names a model they want to use; don't infer it, don't suggest it proactively, don't add it just because it's available. If you do set it, prepend it to both the raw and tmux-wrapped commands.
Read the instructions file for this subcommand from references/:
- setup-project — interactively set up a new AnyPoC project
- find-bugs — pick a scanning strategy and build an
anypoc scan run command
- run-poc — generate PoCs from bug reports produced by an earlier scan (
anypoc poc run)
- retry-poc — rerun PoC generation on an existing bug (new attempt), optionally with extra guidance (
anypoc poc retry)
- hunt — scan for bugs and generate PoCs concurrently in one pass (
anypoc hunt run)
- dashboard — launch the AnyPoC web dashboard (optionally in tmux)
- status — list active anypoc tmux sessions and inspect one for a status summary
- help — list every subcommand with a one-line description
If "$0" does not match any subcommand above, tell the user the available subcommands and stop.
